Efficient Accelerator for Dilated and Transposed Convolution with Decomposition
Kuo-Wei Chang, Tian‐Sheuan Chang
Abstract
Hardware acceleration for dilated and transposed convolution enables real time execution of related tasks like segmentation, but current designs are specific for these convolutional types or suffer from complex control for reconfigurable designs. This paper presents a design that decomposes input or weight for dilated and transposed convolutions respectively to skip redundant computations and thus executes efficiently on existing dense CNN hardware as well. The proposed architecture can cut down 87.8% of the cycle counts to achieve 8.2X speedup over a naive execution for the ENet case.
Topics & Concepts
Computer scienceSpeedupConvolution (computer science)ComputationParallel computingAccelerationHardware accelerationSegmentationComputer hardwareField-programmable gate arrayAlgorithmArtificial intelligenceArtificial neural networkPhysicsClassical mechanicsAdvanced Neural Network ApplicationsCCD and CMOS Imaging SensorsAdvanced Memory and Neural Computing